Beyond the Bot: How Sanoma Media Finland Solved the ‘Miracle Wire’ Crisis to Scale AI Editorial Workflows
A simple USB cable—dubbed the “miracle wire” by frustrated staff—nearly derailed one of Finland’s most ambitious pushes into newsroom automation.
At Sanoma Media Finland, the quest to implement a sophisticated AI editorial workflow hit a surprising wall. It wasn’t a failure of the algorithms or a lack of computing power; it was a failure of the physical process of how journalists handled their phone interviews.
Speaking at the Frankfurt AI Forum, Pauliina Toivanen, Development Manager at Sanoma, revealed that the company’s initial attempts to automate draft articles were stifled by sheer operational chaos.
With a newsroom of roughly 450 journalists—including 200 focused on hard news—Sanoma produces hundreds of interviews weekly. However, there was no unified method for capturing them. Some reporters used digital recorders and manual transfers; others utilized tools like Elisa Ring; some didn’t record at all.
“Our biggest problem was not the AI,” Toivanen noted. “It was everything that happened before it. If there is no standard input, there is no scalable AI.”
The Pivot: Process Over Programming
The realization was stark: you cannot build reliable automation on top of variation. The “miracle wire” was a symptom of a fragmented system that made scaling impossible.
Rather than chasing a more powerful AI, Sanoma shifted its focus to the plumbing of its operation. The team expanded the use of Elisa Ring, a mobile service that records interviews and automatically emails the audio to the journalist.
By eliminating separate devices and manual uploads, Sanoma created a consistent entry point. The result? Immediate adoption. Journalists, tired of the friction of their legacy systems, embraced the simplicity.
Once the input was standardized, the team moved to the AI layer. Developed through the WAN-IFRA GAMI incubator programme in collaboration with Limecraft, the new pipeline handled transcription, summarization, and draft generation in a seamless sequence.
The Struggle to Define ‘Good’
Even with a clean pipeline, Sanoma encountered a philosophical hurdle: how do you quantify “quality” in journalism?
In the Finnish language, where a minor linguistic slip can fundamentally alter the meaning of a sentence, transcription accuracy is a high-stakes game. The team found that defining what constituted a “good” output was actually more difficult than coding the AI itself.
Early drafts often suffered from “hallucinations”—the AI would prioritize the wrong quotes or misorder critical information. This led to a pragmatic shift in definition: “good” was no longer an abstract measure of accuracy, but a measure of the remaining work a journalist had to do after the AI had finished.
Does this mean the journalist is becoming obsolete? Far from it.
The final layer of the workflow remains strictly human. Journalists still decide which conversations are too sensitive to record—such as those involving mental health—and they remain the sole arbiters of what is published. As Toivanen emphasized, the AI provides a starting point, but the journalist provides the insight.
How much of your own professional workflow is currently hindered by “miracle wires”—those outdated, manual workarounds we’ve simply grown used to? Could a shift in process be more valuable than a new piece of software?
For a deeper look at how these strategies are being mirrored globally, the Reuters Institute for the Study of Journalism provides extensive research on the intersection of AI and media ethics.
This journey, detailed in the original WAN-IFRA report, serves as a blueprint for any organization attempting to merge legacy human work with cutting-edge technology.
The Golden Rule of Automation: Standardize, Then Scale
The Sanoma case study offers a timeless lesson for the digital age: AI is an accelerant, not a cure.
If you apply AI to an inefficient process, you simply accelerate the inefficiency. To build a truly scalable system, organizations must adhere to three core principles:
1. The Input Law
The quality of your AI output is a direct reflection of the quality of your input. Whether it is audio files in a newsroom or lead data in a CRM, variation is the enemy of automation. Before investing in LLMs, invest in a “Single Source of Truth” for your data entry.
2. Practical Quality Metrics
Avoid the trap of pursuing 100% theoretical accuracy. Instead, measure “Time to Final Version.” If an AI reduces a journalist’s drafting time from four hours to two, it is a success, even if the first draft requires significant editing.
3. The Human-in-the-Loop (HITL) Necessity
As noted by Nieman Lab, the most successful AI integrations are those that augment human expertise rather than attempting to replace it. The human “editor” is the essential fail-safe against AI hallucinations and ethical lapses.
Frequently Asked Questions About AI Editorial Workflows
What is the biggest challenge in implementing an AI editorial workflow?
The primary challenge is often not the technology itself, but the lack of standardized input. AI cannot scale if the data it receives is inconsistent or gathered through fragmented processes.
How does standardization improve an AI editorial workflow?
Standardization ensures a consistent entry point for data. By removing manual transfers and varying recording methods, an organization creates a reliable pipeline that AI can process without errors.
Can an AI editorial workflow replace journalists?
No. In professional newsrooms, the final layer remains editorial. Journalists are essential for deciding what to record, interpreting AI outputs, and ensuring the ethical handling of sensitive information.
How do you define ‘good’ output in an AI editorial workflow?
In a practical newsroom setting, ‘good’ is defined by the amount of manual correction a journalist must perform after the AI step. Quality is tied to utility and the reduction of human friction.
What tools support a modern AI editorial workflow?
Tools like Elisa Ring for automated recording, combined with transcription and summarization pipelines developed through initiatives like the WAN-IFRA GAMI incubator, can streamline the process.
Join the Conversation: Do you believe the “human-in-the-loop” model is sufficient to prevent AI errors in news, or do we need stricter automated guardrails? Share your thoughts in the comments below and share this article with your network to spark a debate on the future of journalism!
Discover more from Archyworldys
Subscribe to get the latest posts sent to your email.
